Performance Improvement of Human Activity Recognition based on Ensemble Empirical Mode Decomposition (EEMD)

Cell phone and advanced hardware, for example, fitness trackers, heart observing, and wearable gadgets are more regularly used nowadays to capture human exercises. Inertial Measurement Unit (IMU) sensor can read some parameter from human activity. Indicator and position formed from that sensor can be translated back by machine learning to classily human activities. Classification of human exercises known by the term Human Activity Recognition (HAR). Cell phone IMU sensor's data is not linear and stationary. Feature from non-linear signal can be extracted better by using non-linear and non-stationary signal decomposition algorithm than by using conventional frequency analysis (Fourier Transform or Wavelet Transform). Ensemble Empirical Mode Decomposition (EEMD) method is better than Empirical Mode Decomposition (EMD) because EEMD utilize nonlinear signal decomposition based on either time-domain or frequency-domain. For further analysis, multi parameter added from EEMD signal processed with Hilbert-Huang Transform (HHT) to get instantaneous energy density. Instantaneous energy density is representing the absolute amplitude of signal over time and also marginal spectrum. Marginal spectrum shows the amplitude signal in frequency domain. Instantaneous energy density and amplitude of signal becomes selected properties for classification process. The novel approach of this research is joining EEMD process as a raw signal modifier and HHT as feature extraction process. Naïve Bayes, Support Vector Machine (SYUI), and random forest used as machine learning classifier. The highest accuracy obtained from the Random Forest classifier and overall accuracy of three classifiers is 95% for all four performance indexes: recall, precision, F-measure, and accuracy.